1 // Copyright 2012 the V8 project authors. All rights reserved. 2 // Use of this source code is governed by a BSD-style license that can be 3 // found in the LICENSE file. 4 5 #include "src/mips64/codegen-mips64.h" 6 7 #if V8_TARGET_ARCH_MIPS64 8 9 #include <memory> 10 11 #include "src/codegen.h" 12 #include "src/macro-assembler.h" 13 #include "src/mips64/simulator-mips64.h" 14 15 namespace v8 { 16 namespace internal { 17 18 19 #define __ masm. 20 21 22 #if defined(V8_HOST_ARCH_MIPS) 23 MemCopyUint8Function CreateMemCopyUint8Function(Isolate* isolate, 24 MemCopyUint8Function stub) { 25 #if defined(USE_SIMULATOR) 26 return stub; 27 #else 28 29 size_t actual_size; 30 byte* buffer = 31 static_cast<byte*>(base::OS::Allocate(3 * KB, &actual_size, true)); 32 if (buffer == nullptr) return stub; 33 34 // This code assumes that cache lines are 32 bytes and if the cache line is 35 // larger it will not work correctly. 36 MacroAssembler masm(isolate, buffer, static_cast<int>(actual_size), 37 CodeObjectRequired::kNo); 38 39 { 40 Label lastb, unaligned, aligned, chkw, 41 loop16w, chk1w, wordCopy_loop, skip_pref, lastbloop, 42 leave, ua_chk16w, ua_loop16w, ua_skip_pref, ua_chkw, 43 ua_chk1w, ua_wordCopy_loop, ua_smallCopy, ua_smallCopy_loop; 44 45 // The size of each prefetch. 46 uint32_t pref_chunk = 32; 47 // The maximum size of a prefetch, it must not be less than pref_chunk. 48 // If the real size of a prefetch is greater than max_pref_size and 49 // the kPrefHintPrepareForStore hint is used, the code will not work 50 // correctly. 51 uint32_t max_pref_size = 128; 52 DCHECK(pref_chunk < max_pref_size); 53 54 // pref_limit is set based on the fact that we never use an offset 55 // greater then 5 on a store pref and that a single pref can 56 // never be larger then max_pref_size. 57 uint32_t pref_limit = (5 * pref_chunk) + max_pref_size; 58 int32_t pref_hint_load = kPrefHintLoadStreamed; 59 int32_t pref_hint_store = kPrefHintPrepareForStore; 60 uint32_t loadstore_chunk = 4; 61 62 // The initial prefetches may fetch bytes that are before the buffer being 63 // copied. Start copies with an offset of 4 so avoid this situation when 64 // using kPrefHintPrepareForStore. 65 DCHECK(pref_hint_store != kPrefHintPrepareForStore || 66 pref_chunk * 4 >= max_pref_size); 67 // If the size is less than 8, go to lastb. Regardless of size, 68 // copy dst pointer to v0 for the retuen value. 69 __ slti(a6, a2, 2 * loadstore_chunk); 70 __ bne(a6, zero_reg, &lastb); 71 __ mov(v0, a0); // In delay slot. 72 73 // If src and dst have different alignments, go to unaligned, if they 74 // have the same alignment (but are not actually aligned) do a partial 75 // load/store to make them aligned. If they are both already aligned 76 // we can start copying at aligned. 77 __ xor_(t8, a1, a0); 78 __ andi(t8, t8, loadstore_chunk - 1); // t8 is a0/a1 word-displacement. 79 __ bne(t8, zero_reg, &unaligned); 80 __ subu(a3, zero_reg, a0); // In delay slot. 81 82 __ andi(a3, a3, loadstore_chunk - 1); // Copy a3 bytes to align a0/a1. 83 __ beq(a3, zero_reg, &aligned); // Already aligned. 84 __ subu(a2, a2, a3); // In delay slot. a2 is the remining bytes count. 85 86 if (kArchEndian == kLittle) { 87 __ lwr(t8, MemOperand(a1)); 88 __ addu(a1, a1, a3); 89 __ swr(t8, MemOperand(a0)); 90 __ addu(a0, a0, a3); 91 } else { 92 __ lwl(t8, MemOperand(a1)); 93 __ addu(a1, a1, a3); 94 __ swl(t8, MemOperand(a0)); 95 __ addu(a0, a0, a3); 96 } 97 98 // Now dst/src are both aligned to (word) aligned addresses. Set a2 to 99 // count how many bytes we have to copy after all the 64 byte chunks are 100 // copied and a3 to the dst pointer after all the 64 byte chunks have been 101 // copied. We will loop, incrementing a0 and a1 until a0 equals a3. 102 __ bind(&aligned); 103 __ andi(t8, a2, 0x3f); 104 __ beq(a2, t8, &chkw); // Less than 64? 105 __ subu(a3, a2, t8); // In delay slot. 106 __ addu(a3, a0, a3); // Now a3 is the final dst after loop. 107 108 // When in the loop we prefetch with kPrefHintPrepareForStore hint, 109 // in this case the a0+x should be past the "a4-32" address. This means: 110 // for x=128 the last "safe" a0 address is "a4-160". Alternatively, for 111 // x=64 the last "safe" a0 address is "a4-96". In the current version we 112 // will use "pref hint, 128(a0)", so "a4-160" is the limit. 113 if (pref_hint_store == kPrefHintPrepareForStore) { 114 __ addu(a4, a0, a2); // a4 is the "past the end" address. 115 __ Subu(t9, a4, pref_limit); // t9 is the "last safe pref" address. 116 } 117 118 __ Pref(pref_hint_load, MemOperand(a1, 0 * pref_chunk)); 119 __ Pref(pref_hint_load, MemOperand(a1, 1 * pref_chunk)); 120 __ Pref(pref_hint_load, MemOperand(a1, 2 * pref_chunk)); 121 __ Pref(pref_hint_load, MemOperand(a1, 3 * pref_chunk)); 122 123 if (pref_hint_store != kPrefHintPrepareForStore) { 124 __ Pref(pref_hint_store, MemOperand(a0, 1 * pref_chunk)); 125 __ Pref(pref_hint_store, MemOperand(a0, 2 * pref_chunk)); 126 __ Pref(pref_hint_store, MemOperand(a0, 3 * pref_chunk)); 127 } 128 __ bind(&loop16w); 129 __ lw(a4, MemOperand(a1)); 130 131 if (pref_hint_store == kPrefHintPrepareForStore) { 132 __ sltu(v1, t9, a0); // If a0 > t9, don't use next prefetch. 133 __ Branch(USE_DELAY_SLOT, &skip_pref, gt, v1, Operand(zero_reg)); 134 } 135 __ lw(a5, MemOperand(a1, 1, loadstore_chunk)); // Maybe in delay slot. 136 137 __ Pref(pref_hint_store, MemOperand(a0, 4 * pref_chunk)); 138 __ Pref(pref_hint_store, MemOperand(a0, 5 * pref_chunk)); 139 140 __ bind(&skip_pref); 141 __ lw(a6, MemOperand(a1, 2, loadstore_chunk)); 142 __ lw(a7, MemOperand(a1, 3, loadstore_chunk)); 143 __ lw(t0, MemOperand(a1, 4, loadstore_chunk)); 144 __ lw(t1, MemOperand(a1, 5, loadstore_chunk)); 145 __ lw(t2, MemOperand(a1, 6, loadstore_chunk)); 146 __ lw(t3, MemOperand(a1, 7, loadstore_chunk)); 147 __ Pref(pref_hint_load, MemOperand(a1, 4 * pref_chunk)); 148 149 __ sw(a4, MemOperand(a0)); 150 __ sw(a5, MemOperand(a0, 1, loadstore_chunk)); 151 __ sw(a6, MemOperand(a0, 2, loadstore_chunk)); 152 __ sw(a7, MemOperand(a0, 3, loadstore_chunk)); 153 __ sw(t0, MemOperand(a0, 4, loadstore_chunk)); 154 __ sw(t1, MemOperand(a0, 5, loadstore_chunk)); 155 __ sw(t2, MemOperand(a0, 6, loadstore_chunk)); 156 __ sw(t3, MemOperand(a0, 7, loadstore_chunk)); 157 158 __ lw(a4, MemOperand(a1, 8, loadstore_chunk)); 159 __ lw(a5, MemOperand(a1, 9, loadstore_chunk)); 160 __ lw(a6, MemOperand(a1, 10, loadstore_chunk)); 161 __ lw(a7, MemOperand(a1, 11, loadstore_chunk)); 162 __ lw(t0, MemOperand(a1, 12, loadstore_chunk)); 163 __ lw(t1, MemOperand(a1, 13, loadstore_chunk)); 164 __ lw(t2, MemOperand(a1, 14, loadstore_chunk)); 165 __ lw(t3, MemOperand(a1, 15, loadstore_chunk)); 166 __ Pref(pref_hint_load, MemOperand(a1, 5 * pref_chunk)); 167 168 __ sw(a4, MemOperand(a0, 8, loadstore_chunk)); 169 __ sw(a5, MemOperand(a0, 9, loadstore_chunk)); 170 __ sw(a6, MemOperand(a0, 10, loadstore_chunk)); 171 __ sw(a7, MemOperand(a0, 11, loadstore_chunk)); 172 __ sw(t0, MemOperand(a0, 12, loadstore_chunk)); 173 __ sw(t1, MemOperand(a0, 13, loadstore_chunk)); 174 __ sw(t2, MemOperand(a0, 14, loadstore_chunk)); 175 __ sw(t3, MemOperand(a0, 15, loadstore_chunk)); 176 __ addiu(a0, a0, 16 * loadstore_chunk); 177 __ bne(a0, a3, &loop16w); 178 __ addiu(a1, a1, 16 * loadstore_chunk); // In delay slot. 179 __ mov(a2, t8); 180 181 // Here we have src and dest word-aligned but less than 64-bytes to go. 182 // Check for a 32 bytes chunk and copy if there is one. Otherwise jump 183 // down to chk1w to handle the tail end of the copy. 184 __ bind(&chkw); 185 __ Pref(pref_hint_load, MemOperand(a1, 0 * pref_chunk)); 186 __ andi(t8, a2, 0x1f); 187 __ beq(a2, t8, &chk1w); // Less than 32? 188 __ nop(); // In delay slot. 189 __ lw(a4, MemOperand(a1)); 190 __ lw(a5, MemOperand(a1, 1, loadstore_chunk)); 191 __ lw(a6, MemOperand(a1, 2, loadstore_chunk)); 192 __ lw(a7, MemOperand(a1, 3, loadstore_chunk)); 193 __ lw(t0, MemOperand(a1, 4, loadstore_chunk)); 194 __ lw(t1, MemOperand(a1, 5, loadstore_chunk)); 195 __ lw(t2, MemOperand(a1, 6, loadstore_chunk)); 196 __ lw(t3, MemOperand(a1, 7, loadstore_chunk)); 197 __ addiu(a1, a1, 8 * loadstore_chunk); 198 __ sw(a4, MemOperand(a0)); 199 __ sw(a5, MemOperand(a0, 1, loadstore_chunk)); 200 __ sw(a6, MemOperand(a0, 2, loadstore_chunk)); 201 __ sw(a7, MemOperand(a0, 3, loadstore_chunk)); 202 __ sw(t0, MemOperand(a0, 4, loadstore_chunk)); 203 __ sw(t1, MemOperand(a0, 5, loadstore_chunk)); 204 __ sw(t2, MemOperand(a0, 6, loadstore_chunk)); 205 __ sw(t3, MemOperand(a0, 7, loadstore_chunk)); 206 __ addiu(a0, a0, 8 * loadstore_chunk); 207 208 // Here we have less than 32 bytes to copy. Set up for a loop to copy 209 // one word at a time. Set a2 to count how many bytes we have to copy 210 // after all the word chunks are copied and a3 to the dst pointer after 211 // all the word chunks have been copied. We will loop, incrementing a0 212 // and a1 untill a0 equals a3. 213 __ bind(&chk1w); 214 __ andi(a2, t8, loadstore_chunk - 1); 215 __ beq(a2, t8, &lastb); 216 __ subu(a3, t8, a2); // In delay slot. 217 __ addu(a3, a0, a3); 218 219 __ bind(&wordCopy_loop); 220 __ lw(a7, MemOperand(a1)); 221 __ addiu(a0, a0, loadstore_chunk); 222 __ addiu(a1, a1, loadstore_chunk); 223 __ bne(a0, a3, &wordCopy_loop); 224 __ sw(a7, MemOperand(a0, -1, loadstore_chunk)); // In delay slot. 225 226 __ bind(&lastb); 227 __ Branch(&leave, le, a2, Operand(zero_reg)); 228 __ addu(a3, a0, a2); 229 230 __ bind(&lastbloop); 231 __ lb(v1, MemOperand(a1)); 232 __ addiu(a0, a0, 1); 233 __ addiu(a1, a1, 1); 234 __ bne(a0, a3, &lastbloop); 235 __ sb(v1, MemOperand(a0, -1)); // In delay slot. 236 237 __ bind(&leave); 238 __ jr(ra); 239 __ nop(); 240 241 // Unaligned case. Only the dst gets aligned so we need to do partial 242 // loads of the source followed by normal stores to the dst (once we 243 // have aligned the destination). 244 __ bind(&unaligned); 245 __ andi(a3, a3, loadstore_chunk - 1); // Copy a3 bytes to align a0/a1. 246 __ beq(a3, zero_reg, &ua_chk16w); 247 __ subu(a2, a2, a3); // In delay slot. 248 249 if (kArchEndian == kLittle) { 250 __ lwr(v1, MemOperand(a1)); 251 __ lwl(v1, 252 MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one)); 253 __ addu(a1, a1, a3); 254 __ swr(v1, MemOperand(a0)); 255 __ addu(a0, a0, a3); 256 } else { 257 __ lwl(v1, MemOperand(a1)); 258 __ lwr(v1, 259 MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one)); 260 __ addu(a1, a1, a3); 261 __ swl(v1, MemOperand(a0)); 262 __ addu(a0, a0, a3); 263 } 264 265 // Now the dst (but not the source) is aligned. Set a2 to count how many 266 // bytes we have to copy after all the 64 byte chunks are copied and a3 to 267 // the dst pointer after all the 64 byte chunks have been copied. We will 268 // loop, incrementing a0 and a1 until a0 equals a3. 269 __ bind(&ua_chk16w); 270 __ andi(t8, a2, 0x3f); 271 __ beq(a2, t8, &ua_chkw); 272 __ subu(a3, a2, t8); // In delay slot. 273 __ addu(a3, a0, a3); 274 275 if (pref_hint_store == kPrefHintPrepareForStore) { 276 __ addu(a4, a0, a2); 277 __ Subu(t9, a4, pref_limit); 278 } 279 280 __ Pref(pref_hint_load, MemOperand(a1, 0 * pref_chunk)); 281 __ Pref(pref_hint_load, MemOperand(a1, 1 * pref_chunk)); 282 __ Pref(pref_hint_load, MemOperand(a1, 2 * pref_chunk)); 283 284 if (pref_hint_store != kPrefHintPrepareForStore) { 285 __ Pref(pref_hint_store, MemOperand(a0, 1 * pref_chunk)); 286 __ Pref(pref_hint_store, MemOperand(a0, 2 * pref_chunk)); 287 __ Pref(pref_hint_store, MemOperand(a0, 3 * pref_chunk)); 288 } 289 290 __ bind(&ua_loop16w); 291 if (kArchEndian == kLittle) { 292 __ Pref(pref_hint_load, MemOperand(a1, 3 * pref_chunk)); 293 __ lwr(a4, MemOperand(a1)); 294 __ lwr(a5, MemOperand(a1, 1, loadstore_chunk)); 295 __ lwr(a6, MemOperand(a1, 2, loadstore_chunk)); 296 297 if (pref_hint_store == kPrefHintPrepareForStore) { 298 __ sltu(v1, t9, a0); 299 __ Branch(USE_DELAY_SLOT, &ua_skip_pref, gt, v1, Operand(zero_reg)); 300 } 301 __ lwr(a7, MemOperand(a1, 3, loadstore_chunk)); // Maybe in delay slot. 302 303 __ Pref(pref_hint_store, MemOperand(a0, 4 * pref_chunk)); 304 __ Pref(pref_hint_store, MemOperand(a0, 5 * pref_chunk)); 305 306 __ bind(&ua_skip_pref); 307 __ lwr(t0, MemOperand(a1, 4, loadstore_chunk)); 308 __ lwr(t1, MemOperand(a1, 5, loadstore_chunk)); 309 __ lwr(t2, MemOperand(a1, 6, loadstore_chunk)); 310 __ lwr(t3, MemOperand(a1, 7, loadstore_chunk)); 311 __ lwl(a4, 312 MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one)); 313 __ lwl(a5, 314 MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one)); 315 __ lwl(a6, 316 MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one)); 317 __ lwl(a7, 318 MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one)); 319 __ lwl(t0, 320 MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one)); 321 __ lwl(t1, 322 MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one)); 323 __ lwl(t2, 324 MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one)); 325 __ lwl(t3, 326 MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one)); 327 } else { 328 __ Pref(pref_hint_load, MemOperand(a1, 3 * pref_chunk)); 329 __ lwl(a4, MemOperand(a1)); 330 __ lwl(a5, MemOperand(a1, 1, loadstore_chunk)); 331 __ lwl(a6, MemOperand(a1, 2, loadstore_chunk)); 332 333 if (pref_hint_store == kPrefHintPrepareForStore) { 334 __ sltu(v1, t9, a0); 335 __ Branch(USE_DELAY_SLOT, &ua_skip_pref, gt, v1, Operand(zero_reg)); 336 } 337 __ lwl(a7, MemOperand(a1, 3, loadstore_chunk)); // Maybe in delay slot. 338 339 __ Pref(pref_hint_store, MemOperand(a0, 4 * pref_chunk)); 340 __ Pref(pref_hint_store, MemOperand(a0, 5 * pref_chunk)); 341 342 __ bind(&ua_skip_pref); 343 __ lwl(t0, MemOperand(a1, 4, loadstore_chunk)); 344 __ lwl(t1, MemOperand(a1, 5, loadstore_chunk)); 345 __ lwl(t2, MemOperand(a1, 6, loadstore_chunk)); 346 __ lwl(t3, MemOperand(a1, 7, loadstore_chunk)); 347 __ lwr(a4, 348 MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one)); 349 __ lwr(a5, 350 MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one)); 351 __ lwr(a6, 352 MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one)); 353 __ lwr(a7, 354 MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one)); 355 __ lwr(t0, 356 MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one)); 357 __ lwr(t1, 358 MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one)); 359 __ lwr(t2, 360 MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one)); 361 __ lwr(t3, 362 MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one)); 363 } 364 __ Pref(pref_hint_load, MemOperand(a1, 4 * pref_chunk)); 365 __ sw(a4, MemOperand(a0)); 366 __ sw(a5, MemOperand(a0, 1, loadstore_chunk)); 367 __ sw(a6, MemOperand(a0, 2, loadstore_chunk)); 368 __ sw(a7, MemOperand(a0, 3, loadstore_chunk)); 369 __ sw(t0, MemOperand(a0, 4, loadstore_chunk)); 370 __ sw(t1, MemOperand(a0, 5, loadstore_chunk)); 371 __ sw(t2, MemOperand(a0, 6, loadstore_chunk)); 372 __ sw(t3, MemOperand(a0, 7, loadstore_chunk)); 373 if (kArchEndian == kLittle) { 374 __ lwr(a4, MemOperand(a1, 8, loadstore_chunk)); 375 __ lwr(a5, MemOperand(a1, 9, loadstore_chunk)); 376 __ lwr(a6, MemOperand(a1, 10, loadstore_chunk)); 377 __ lwr(a7, MemOperand(a1, 11, loadstore_chunk)); 378 __ lwr(t0, MemOperand(a1, 12, loadstore_chunk)); 379 __ lwr(t1, MemOperand(a1, 13, loadstore_chunk)); 380 __ lwr(t2, MemOperand(a1, 14, loadstore_chunk)); 381 __ lwr(t3, MemOperand(a1, 15, loadstore_chunk)); 382 __ lwl(a4, 383 MemOperand(a1, 9, loadstore_chunk, MemOperand::offset_minus_one)); 384 __ lwl(a5, 385 MemOperand(a1, 10, loadstore_chunk, MemOperand::offset_minus_one)); 386 __ lwl(a6, 387 MemOperand(a1, 11, loadstore_chunk, MemOperand::offset_minus_one)); 388 __ lwl(a7, 389 MemOperand(a1, 12, loadstore_chunk, MemOperand::offset_minus_one)); 390 __ lwl(t0, 391 MemOperand(a1, 13, loadstore_chunk, MemOperand::offset_minus_one)); 392 __ lwl(t1, 393 MemOperand(a1, 14, loadstore_chunk, MemOperand::offset_minus_one)); 394 __ lwl(t2, 395 MemOperand(a1, 15, loadstore_chunk, MemOperand::offset_minus_one)); 396 __ lwl(t3, 397 MemOperand(a1, 16, loadstore_chunk, MemOperand::offset_minus_one)); 398 } else { 399 __ lwl(a4, MemOperand(a1, 8, loadstore_chunk)); 400 __ lwl(a5, MemOperand(a1, 9, loadstore_chunk)); 401 __ lwl(a6, MemOperand(a1, 10, loadstore_chunk)); 402 __ lwl(a7, MemOperand(a1, 11, loadstore_chunk)); 403 __ lwl(t0, MemOperand(a1, 12, loadstore_chunk)); 404 __ lwl(t1, MemOperand(a1, 13, loadstore_chunk)); 405 __ lwl(t2, MemOperand(a1, 14, loadstore_chunk)); 406 __ lwl(t3, MemOperand(a1, 15, loadstore_chunk)); 407 __ lwr(a4, 408 MemOperand(a1, 9, loadstore_chunk, MemOperand::offset_minus_one)); 409 __ lwr(a5, 410 MemOperand(a1, 10, loadstore_chunk, MemOperand::offset_minus_one)); 411 __ lwr(a6, 412 MemOperand(a1, 11, loadstore_chunk, MemOperand::offset_minus_one)); 413 __ lwr(a7, 414 MemOperand(a1, 12, loadstore_chunk, MemOperand::offset_minus_one)); 415 __ lwr(t0, 416 MemOperand(a1, 13, loadstore_chunk, MemOperand::offset_minus_one)); 417 __ lwr(t1, 418 MemOperand(a1, 14, loadstore_chunk, MemOperand::offset_minus_one)); 419 __ lwr(t2, 420 MemOperand(a1, 15, loadstore_chunk, MemOperand::offset_minus_one)); 421 __ lwr(t3, 422 MemOperand(a1, 16, loadstore_chunk, MemOperand::offset_minus_one)); 423 } 424 __ Pref(pref_hint_load, MemOperand(a1, 5 * pref_chunk)); 425 __ sw(a4, MemOperand(a0, 8, loadstore_chunk)); 426 __ sw(a5, MemOperand(a0, 9, loadstore_chunk)); 427 __ sw(a6, MemOperand(a0, 10, loadstore_chunk)); 428 __ sw(a7, MemOperand(a0, 11, loadstore_chunk)); 429 __ sw(t0, MemOperand(a0, 12, loadstore_chunk)); 430 __ sw(t1, MemOperand(a0, 13, loadstore_chunk)); 431 __ sw(t2, MemOperand(a0, 14, loadstore_chunk)); 432 __ sw(t3, MemOperand(a0, 15, loadstore_chunk)); 433 __ addiu(a0, a0, 16 * loadstore_chunk); 434 __ bne(a0, a3, &ua_loop16w); 435 __ addiu(a1, a1, 16 * loadstore_chunk); // In delay slot. 436 __ mov(a2, t8); 437 438 // Here less than 64-bytes. Check for 439 // a 32 byte chunk and copy if there is one. Otherwise jump down to 440 // ua_chk1w to handle the tail end of the copy. 441 __ bind(&ua_chkw); 442 __ Pref(pref_hint_load, MemOperand(a1)); 443 __ andi(t8, a2, 0x1f); 444 445 __ beq(a2, t8, &ua_chk1w); 446 __ nop(); // In delay slot. 447 if (kArchEndian == kLittle) { 448 __ lwr(a4, MemOperand(a1)); 449 __ lwr(a5, MemOperand(a1, 1, loadstore_chunk)); 450 __ lwr(a6, MemOperand(a1, 2, loadstore_chunk)); 451 __ lwr(a7, MemOperand(a1, 3, loadstore_chunk)); 452 __ lwr(t0, MemOperand(a1, 4, loadstore_chunk)); 453 __ lwr(t1, MemOperand(a1, 5, loadstore_chunk)); 454 __ lwr(t2, MemOperand(a1, 6, loadstore_chunk)); 455 __ lwr(t3, MemOperand(a1, 7, loadstore_chunk)); 456 __ lwl(a4, 457 MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one)); 458 __ lwl(a5, 459 MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one)); 460 __ lwl(a6, 461 MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one)); 462 __ lwl(a7, 463 MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one)); 464 __ lwl(t0, 465 MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one)); 466 __ lwl(t1, 467 MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one)); 468 __ lwl(t2, 469 MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one)); 470 __ lwl(t3, 471 MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one)); 472 } else { 473 __ lwl(a4, MemOperand(a1)); 474 __ lwl(a5, MemOperand(a1, 1, loadstore_chunk)); 475 __ lwl(a6, MemOperand(a1, 2, loadstore_chunk)); 476 __ lwl(a7, MemOperand(a1, 3, loadstore_chunk)); 477 __ lwl(t0, MemOperand(a1, 4, loadstore_chunk)); 478 __ lwl(t1, MemOperand(a1, 5, loadstore_chunk)); 479 __ lwl(t2, MemOperand(a1, 6, loadstore_chunk)); 480 __ lwl(t3, MemOperand(a1, 7, loadstore_chunk)); 481 __ lwr(a4, 482 MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one)); 483 __ lwr(a5, 484 MemOperand(a1, 2, loadstore_chunk, MemOperand::offset_minus_one)); 485 __ lwr(a6, 486 MemOperand(a1, 3, loadstore_chunk, MemOperand::offset_minus_one)); 487 __ lwr(a7, 488 MemOperand(a1, 4, loadstore_chunk, MemOperand::offset_minus_one)); 489 __ lwr(t0, 490 MemOperand(a1, 5, loadstore_chunk, MemOperand::offset_minus_one)); 491 __ lwr(t1, 492 MemOperand(a1, 6, loadstore_chunk, MemOperand::offset_minus_one)); 493 __ lwr(t2, 494 MemOperand(a1, 7, loadstore_chunk, MemOperand::offset_minus_one)); 495 __ lwr(t3, 496 MemOperand(a1, 8, loadstore_chunk, MemOperand::offset_minus_one)); 497 } 498 __ addiu(a1, a1, 8 * loadstore_chunk); 499 __ sw(a4, MemOperand(a0)); 500 __ sw(a5, MemOperand(a0, 1, loadstore_chunk)); 501 __ sw(a6, MemOperand(a0, 2, loadstore_chunk)); 502 __ sw(a7, MemOperand(a0, 3, loadstore_chunk)); 503 __ sw(t0, MemOperand(a0, 4, loadstore_chunk)); 504 __ sw(t1, MemOperand(a0, 5, loadstore_chunk)); 505 __ sw(t2, MemOperand(a0, 6, loadstore_chunk)); 506 __ sw(t3, MemOperand(a0, 7, loadstore_chunk)); 507 __ addiu(a0, a0, 8 * loadstore_chunk); 508 509 // Less than 32 bytes to copy. Set up for a loop to 510 // copy one word at a time. 511 __ bind(&ua_chk1w); 512 __ andi(a2, t8, loadstore_chunk - 1); 513 __ beq(a2, t8, &ua_smallCopy); 514 __ subu(a3, t8, a2); // In delay slot. 515 __ addu(a3, a0, a3); 516 517 __ bind(&ua_wordCopy_loop); 518 if (kArchEndian == kLittle) { 519 __ lwr(v1, MemOperand(a1)); 520 __ lwl(v1, 521 MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one)); 522 } else { 523 __ lwl(v1, MemOperand(a1)); 524 __ lwr(v1, 525 MemOperand(a1, 1, loadstore_chunk, MemOperand::offset_minus_one)); 526 } 527 __ addiu(a0, a0, loadstore_chunk); 528 __ addiu(a1, a1, loadstore_chunk); 529 __ bne(a0, a3, &ua_wordCopy_loop); 530 __ sw(v1, MemOperand(a0, -1, loadstore_chunk)); // In delay slot. 531 532 // Copy the last 8 bytes. 533 __ bind(&ua_smallCopy); 534 __ beq(a2, zero_reg, &leave); 535 __ addu(a3, a0, a2); // In delay slot. 536 537 __ bind(&ua_smallCopy_loop); 538 __ lb(v1, MemOperand(a1)); 539 __ addiu(a0, a0, 1); 540 __ addiu(a1, a1, 1); 541 __ bne(a0, a3, &ua_smallCopy_loop); 542 __ sb(v1, MemOperand(a0, -1)); // In delay slot. 543 544 __ jr(ra); 545 __ nop(); 546 } 547 CodeDesc desc; 548 masm.GetCode(&desc); 549 DCHECK(!RelocInfo::RequiresRelocation(desc)); 550 551 Assembler::FlushICache(isolate, buffer, actual_size); 552 base::OS::ProtectCode(buffer, actual_size); 553 return FUNCTION_CAST<MemCopyUint8Function>(buffer); 554 #endif 555 } 556 #endif 557 558 UnaryMathFunctionWithIsolate CreateSqrtFunction(Isolate* isolate) { 559 #if defined(USE_SIMULATOR) 560 return nullptr; 561 #else 562 size_t actual_size; 563 byte* buffer = 564 static_cast<byte*>(base::OS::Allocate(1 * KB, &actual_size, true)); 565 if (buffer == nullptr) return nullptr; 566 567 MacroAssembler masm(isolate, buffer, static_cast<int>(actual_size), 568 CodeObjectRequired::kNo); 569 570 __ MovFromFloatParameter(f12); 571 __ sqrt_d(f0, f12); 572 __ MovToFloatResult(f0); 573 __ Ret(); 574 575 CodeDesc desc; 576 masm.GetCode(&desc); 577 DCHECK(!RelocInfo::RequiresRelocation(desc)); 578 579 Assembler::FlushICache(isolate, buffer, actual_size); 580 base::OS::ProtectCode(buffer, actual_size); 581 return FUNCTION_CAST<UnaryMathFunctionWithIsolate>(buffer); 582 #endif 583 } 584 585 #undef __ 586 587 588 // ------------------------------------------------------------------------- 589 // Platform-specific RuntimeCallHelper functions. 590 591 void StubRuntimeCallHelper::BeforeCall(MacroAssembler* masm) const { 592 masm->EnterFrame(StackFrame::INTERNAL); 593 DCHECK(!masm->has_frame()); 594 masm->set_has_frame(true); 595 } 596 597 598 void StubRuntimeCallHelper::AfterCall(MacroAssembler* masm) const { 599 masm->LeaveFrame(StackFrame::INTERNAL); 600 DCHECK(masm->has_frame()); 601 masm->set_has_frame(false); 602 } 603 604 605 // ------------------------------------------------------------------------- 606 // Code generators 607 608 #define __ ACCESS_MASM(masm) 609 610 void StringCharLoadGenerator::Generate(MacroAssembler* masm, 611 Register string, 612 Register index, 613 Register result, 614 Label* call_runtime) { 615 Label indirect_string_loaded; 616 __ bind(&indirect_string_loaded); 617 618 // Fetch the instance type of the receiver into result register. 619 __ ld(result, FieldMemOperand(string, HeapObject::kMapOffset)); 620 __ lbu(result, FieldMemOperand(result, Map::kInstanceTypeOffset)); 621 622 // We need special handling for indirect strings. 623 Label check_sequential; 624 __ And(at, result, Operand(kIsIndirectStringMask)); 625 __ Branch(&check_sequential, eq, at, Operand(zero_reg)); 626 627 // Dispatch on the indirect string shape: slice or cons. 628 Label cons_string, thin_string; 629 __ And(at, result, Operand(kStringRepresentationMask)); 630 __ Branch(&cons_string, eq, at, Operand(kConsStringTag)); 631 __ Branch(&thin_string, eq, at, Operand(kThinStringTag)); 632 633 // Handle slices. 634 __ ld(result, FieldMemOperand(string, SlicedString::kOffsetOffset)); 635 __ ld(string, FieldMemOperand(string, SlicedString::kParentOffset)); 636 __ dsra32(at, result, 0); 637 __ Daddu(index, index, at); 638 __ jmp(&indirect_string_loaded); 639 640 // Handle thin strings. 641 __ bind(&thin_string); 642 __ ld(string, FieldMemOperand(string, ThinString::kActualOffset)); 643 __ jmp(&indirect_string_loaded); 644 645 // Handle cons strings. 646 // Check whether the right hand side is the empty string (i.e. if 647 // this is really a flat string in a cons string). If that is not 648 // the case we would rather go to the runtime system now to flatten 649 // the string. 650 __ bind(&cons_string); 651 __ ld(result, FieldMemOperand(string, ConsString::kSecondOffset)); 652 __ LoadRoot(at, Heap::kempty_stringRootIndex); 653 __ Branch(call_runtime, ne, result, Operand(at)); 654 // Get the first of the two strings and load its instance type. 655 __ ld(string, FieldMemOperand(string, ConsString::kFirstOffset)); 656 __ jmp(&indirect_string_loaded); 657 658 // Distinguish sequential and external strings. Only these two string 659 // representations can reach here (slices and flat cons strings have been 660 // reduced to the underlying sequential or external string). 661 Label external_string, check_encoding; 662 __ bind(&check_sequential); 663 STATIC_ASSERT(kSeqStringTag == 0); 664 __ And(at, result, Operand(kStringRepresentationMask)); 665 __ Branch(&external_string, ne, at, Operand(zero_reg)); 666 667 // Prepare sequential strings 668 STATIC_ASSERT(SeqTwoByteString::kHeaderSize == SeqOneByteString::kHeaderSize); 669 __ Daddu(string, 670 string, 671 SeqTwoByteString::kHeaderSize - kHeapObjectTag); 672 __ jmp(&check_encoding); 673 674 // Handle external strings. 675 __ bind(&external_string); 676 if (FLAG_debug_code) { 677 // Assert that we do not have a cons or slice (indirect strings) here. 678 // Sequential strings have already been ruled out. 679 __ And(at, result, Operand(kIsIndirectStringMask)); 680 __ Assert(eq, kExternalStringExpectedButNotFound, 681 at, Operand(zero_reg)); 682 } 683 // Rule out short external strings. 684 STATIC_ASSERT(kShortExternalStringTag != 0); 685 __ And(at, result, Operand(kShortExternalStringMask)); 686 __ Branch(call_runtime, ne, at, Operand(zero_reg)); 687 __ ld(string, FieldMemOperand(string, ExternalString::kResourceDataOffset)); 688 689 Label one_byte, done; 690 __ bind(&check_encoding); 691 STATIC_ASSERT(kTwoByteStringTag == 0); 692 __ And(at, result, Operand(kStringEncodingMask)); 693 __ Branch(&one_byte, ne, at, Operand(zero_reg)); 694 // Two-byte string. 695 __ Dlsa(at, string, index, 1); 696 __ lhu(result, MemOperand(at)); 697 __ jmp(&done); 698 __ bind(&one_byte); 699 // One_byte string. 700 __ Daddu(at, string, index); 701 __ lbu(result, MemOperand(at)); 702 __ bind(&done); 703 } 704 705 #ifdef DEBUG 706 // nop(CODE_AGE_MARKER_NOP) 707 static const uint32_t kCodeAgePatchFirstInstruction = 0x00010180; 708 #endif 709 710 711 CodeAgingHelper::CodeAgingHelper(Isolate* isolate) { 712 USE(isolate); 713 DCHECK(young_sequence_.length() == kNoCodeAgeSequenceLength); 714 // Since patcher is a large object, allocate it dynamically when needed, 715 // to avoid overloading the stack in stress conditions. 716 // DONT_FLUSH is used because the CodeAgingHelper is initialized early in 717 // the process, before MIPS simulator ICache is setup. 718 std::unique_ptr<CodePatcher> patcher( 719 new CodePatcher(isolate, young_sequence_.start(), 720 young_sequence_.length() / Assembler::kInstrSize, 721 CodePatcher::DONT_FLUSH)); 722 PredictableCodeSizeScope scope(patcher->masm(), young_sequence_.length()); 723 patcher->masm()->PushStandardFrame(a1); 724 patcher->masm()->nop(Assembler::CODE_AGE_SEQUENCE_NOP); 725 patcher->masm()->nop(Assembler::CODE_AGE_SEQUENCE_NOP); 726 patcher->masm()->nop(Assembler::CODE_AGE_SEQUENCE_NOP); 727 } 728 729 730 #ifdef DEBUG 731 bool CodeAgingHelper::IsOld(byte* candidate) const { 732 return Memory::uint32_at(candidate) == kCodeAgePatchFirstInstruction; 733 } 734 #endif 735 736 737 bool Code::IsYoungSequence(Isolate* isolate, byte* sequence) { 738 bool result = isolate->code_aging_helper()->IsYoung(sequence); 739 DCHECK(result || isolate->code_aging_helper()->IsOld(sequence)); 740 return result; 741 } 742 743 Code::Age Code::GetCodeAge(Isolate* isolate, byte* sequence) { 744 if (IsYoungSequence(isolate, sequence)) return kNoAgeCodeAge; 745 746 Address target_address = 747 Assembler::target_address_at(sequence + Assembler::kInstrSize); 748 Code* stub = GetCodeFromTargetAddress(target_address); 749 return GetAgeOfCodeAgeStub(stub); 750 } 751 752 void Code::PatchPlatformCodeAge(Isolate* isolate, byte* sequence, 753 Code::Age age) { 754 uint32_t young_length = isolate->code_aging_helper()->young_sequence_length(); 755 if (age == kNoAgeCodeAge) { 756 isolate->code_aging_helper()->CopyYoungSequenceTo(sequence); 757 Assembler::FlushICache(isolate, sequence, young_length); 758 } else { 759 Code* stub = GetCodeAgeStub(isolate, age); 760 CodePatcher patcher(isolate, sequence, 761 young_length / Assembler::kInstrSize); 762 // Mark this code sequence for FindPlatformCodeAgeSequence(). 763 patcher.masm()->nop(Assembler::CODE_AGE_MARKER_NOP); 764 // Load the stub address to t9 and call it, 765 // GetCodeAge() extracts the stub address from this instruction. 766 patcher.masm()->li( 767 t9, 768 Operand(reinterpret_cast<uint64_t>(stub->instruction_start())), 769 ADDRESS_LOAD); 770 patcher.masm()->nop(); // Prevent jalr to jal optimization. 771 patcher.masm()->jalr(t9, a0); 772 patcher.masm()->nop(); // Branch delay slot nop. 773 patcher.masm()->nop(); // Pad the empty space. 774 } 775 } 776 777 778 #undef __ 779 780 } // namespace internal 781 } // namespace v8 782 783 #endif // V8_TARGET_ARCH_MIPS64 784
